Environ. Eng. Res. 2018

Research Article https://doi.org/10.4491/eer.2017.098 pISSN 1226-1025 eISSN 2005-968X In Press, Uncorrected Proof

Optimization of chemical cleaning for reverse osmosis membranes with organic fouling using statistical design tools

Ki-Bum Park1, Changkyoo Choi1, Hye-Weon Yu1, So-Ryong Chae2, In S. Kim1†

1Global Desalination Research Center (GDRC), School of Earth Sciences and Environmental Engineering, Gwangju Institute of Science and Technology (GIST), Gwangju 61005, Republic of Korea 2Department of Biomedical, Chemical, and Environmental Engineering, University of Cincinnati, 701 Engineering Research Center, Cincinnati, OH 45221, U.S.A.

Abstract The cleaning efficiency of reverse osmosis (RO) membranes inevitably fouled by organic foulants depends upon both chemical (type of cleaning agent, concentration of cleaning solution) and physical (cleaning time, flowrate, temperature) parameters. In attempting to determine the optimal procedures for chemical cleaning organic-fouled RO membranes, the design of experiments concept was employed to evaluate key factors and to predict the flux recovery rate (FRR) after chemical cleaning. From experimental results and based on the predicted FRR of cleaning obtained using the Central Composite Design of Minitab 17, a modified regression model equation was established to explain the chemical cleaning efficiency; the resultant regression coefficient (R2) and adjusted R2 were 83.95% and 76.82%, respectively. Then, using the optimized conditions of chemical cleaning derived from the response optimizer tool (cleaning with 0.68 wt% disodium ethylenediaminetetraacetic acid for 20 min at 20°C with a flowrate of 409 mL/min), a flux recovery of 86.6% was expected. Overall, the results obtained by these experiments confirmed that the equation was adequate for predicting the chemical cleaning efficiency with regards to organic membrane fouling. Keywords: Chemical cleaning, Design of experiment, Organic fouling, Reverse osmosis membrane

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial Li- cense (http://creativecommons.org/licenses/by-nc/3.0/)

which permits unrestricted non-commercial use, distribution, and repro- duction in any medium, provided the original work is properly cited.

Received July 28, 2017 Accepted May 3, 2018 † Corresponding Author E-mail: iskim@gist.ac.kr Tel: +82-62-715-2436 Fax: +82-62-715-2434

ORCID: 0000-0003-4195-5412

Copyright © 2018 Korean Society of Environmental Engineers http://eeer.org

1. Introduction

Reverse osmosis (RO) membrane-based technologies are now widely being used for

producing potable and ultra-pure water from wastewater, seawater, and surface water rather

than using conventional water treatment processes because the energy consumption of the RO

process is much lower and the operational safety is higher [1-4].

However, fouling is inevitable in RO processes, which causes a flux reduction, high energy

demand, and increases membrane damage and the need for replacement. As such, the

productivity and lifetime of RO membranes are severely limited by the fouling [5-7]. For this

reason, periodic membrane cleaning is an integral part of the successful operation of RO

processes. There are two general methods for the removal of foulants: physical cleaning and

chemical cleaning. Physical cleaning includes backflushing, vibration, air sparging, automatic

sponge ball cleaning, and ultra-sonication [8]. Foulants can also be efficiently eliminated

using various chemical cleaning agents [9-10], though the cleaning efficiency is strongly

influenced by parameters such as temperature, cleaning agent concentration, operating

pressure, flowrate, and cleaning time [11].

The operational cost of RO processes can increase due to excessive and frequent

membrane chemical cleaning, with membrane cleaning generally accounting for 50% of the

total operation costs of seawater desalination plants. Frequent membrane chemical cleaning

can also shorten the membrane lifetime [12]; hence, optimized protocols for membrane

chemical cleaning must be employed to overcome the drawbacks of RO membrane operation

[13].

A full factorial design (FFD) allows multiple factors to be investigated at the same time

and enables the more accurate identification of interactions between factors. A central

composite design (CCD) is an alternative model for determining the curvature in quadratic

1

terms. Hence, a design of experiments (DOE) concepts, which is one of response surface

methods that seek the relationships between several descriptive variables and response

variables, was employed in the study of RO membrane cleaning in order to reduce the number

of experimental runs required for determining the effect of changing one factor and thereby

optimizing conditions by balancing tradeoffs. Shams et al. [14] revealed that FFD and CCD

can be used to find the optimal conditions for the filtrate production of a single 8-inch

commercial ultrafiltration membrane with regards to feed pressure, backwash pressure,

forward filtration time, and backwash time. In addition, Yi et al. [15] revealed that FFD can be

employed to screen significant factors affecting the flux decline of anionic polyacrylamide in

ultrafiltration processes. Chen et al. [16] conducted experiments to accurately identify key

factors and their interactions in physical and chemical cleanings based on a statistical factorial

design.

Sodium alginate was selected as the model foulant to simulate organic fouling on the

surface of RO membranes in the presence of Ca2+ ions [17-19]. Using the factorial design

method, the subsets of key factors identified and interactions that impacted the flux recovery

and cleaning efficiency were sequentially used in the CCD to model the response variables

with curvature, and a response surface designed experiment was then used to determine the

optimal setting for each factor during RO membrane cleaning. The objective of this study was

to suggest a methodology for selecting an optimal combination of cleaning agents and

chemical/physical conditions for efficient cleaning, using DOE as a statistical design tool.

Overall, four key chemical/physical factors impacting the RO membrane cleaning efficiency,

as reported in Ang et al. [20] and Garcia-Fayos et al. [11], were selected for use in evaluating

multiple factors set at various levels: cleaning agent type, chemical concentration, cleaning

time, flowrate, and cleaning temperature.

2

Due to the variety of cleaning protocols of commercial membranes by manufacturers,

optimal conditions of membrane cleaning is difficult to be defined. To figure out the optimal

conditions on various membranes, the experiments on membranes for the fouling and

cleaning should be accomplished. Hence many researchers have been studying to find out the

optimal conditions by an experiment [11, 20]. However, too many variables must be

considered to increase the accuracy of the optimal conditions, then, the experiment stage

becomes so complicated. So, this research suggests the procedure for finding the optimal

condition of reverse osmosis membrane cleaning using DOE method to simplify the

experimental stage.

Therefore, this paper discusses differences in the chemical cleaning efficiency between

statistically determined optimal chemical cleaning factors obtained by FFD and CCD

composite and chemical cleaning factors obtained from membrane manufacturer guidelines.

We then confirm whether the DOE method can effectively predict the operational factors for

chemical cleaning.

2. Experimental Methods

2.1. RO Membrane

A thin-film polyamide brackish water RO (BWRO) membrane, RE8040-BLR (Toray

Chemical Korea, Inc., Korea), was used in this study. The permeate flow rate was determined

to be 38.08 L/m2·h (LMH) and stabilized salt rejection was 99.6% under the following

standard manufacturer-recommended test conditions: NaCl solution of 1,500 mg/L, pressure

of 150 psi (10.34 bar) and temperature of 25°C. BWRO membranes were stored in deionized

(DI) water at 4°C prior to each experiment.

3

2.2. Organic Fouling Matter

Sodium alginate (Sigma-Aldrich, USA), extracted from brown algae, was used as the organic

foulant. An organic foulant solution (1 g/L) was prepared by dissolving sodium alginate in DI

water and adding 1 mM CaCl2·2H2O (OCI Company Ltd., Korea) as divalent cations to

aggressively form an organic fouling layer on the RO membranes. Sodium alginate was

stirred for over 24 h to completely dissolve the foulant.

2.3. Cleaning Chemical Agents

Sodium hydroxide (NaOH) from OCI Company Ltd. (Korea), citric acid, sodium dodecyl

sulfate (SDS), and ethylenediaminetetraacetic acid (EDTA) from Sigma-Aldrich (USA) were

used as the chemical cleaning agents.

2.4. Dead-end RO System and Operating Conditions

A lab-scale dead-end filtration system was used to investigate the efficiency of static cleaning

for each of the chemical cleaning agents (i.e., NaOH, SDS, EDTA, and citric acid). Prior to

cleaning test, dense layer of organic fouling was made by dead-end filtration. In Fig. 1, this

filtration system is seen to be comprised of pressure gas, a feed solution reservoir and stirred

hot plate, a permeate tank with digital balance, a membrane cell, a pressure meter, and a data-

acquisition computer. The water flux (Jw; LMH) was measured using a digital balance (GF-

6100, A&D, USA) and was automatically recorded on a computer. The operating conditions

of the dead-end filtration are summarized in Table 1.

2.5. Cross-flow RO System and Operating Conditions

To determine the optimum membrane cleaning conditions, a lab-scale RO system in cross-

4

flow mode was applied. In Fig. 2, the feed and cleaning solutions were circulated using gear

pumps (Micropump, Cole-Parmer, USA). The hydraulic pressure and flowrate were adjusted

using a bypass and pressurization valves. The permeate volume was measured using a digital

balance, and the water flux was calculated using a data-acquisition computer. The operating

conditions of this system are summarized in Table 2.

2.6. Experimental Design Using FFD

FFD has been used to accurately identify interactions between factors and to determine the

effects of one factor from several levels; it can reduce the number of experiments required to

determine the effects derived from the one factor change [16]. The FFD analyzes the

influence of every factor at low (-1), center (0), and high (+1) levels. The specific

experimental parameters are shown in Table 3. In terms of cleaning solution, only EDTA was

used, according to the experimental results of static cleaning.

Detailed levels at low, center and high levels were determined based on membrane

manufacturer guidelines. The curvature effect was investigated while the center point (Ct-Pt)

was added to the FFD. Experimental designs of FFD are presented in Table 4. Factorial plots

were obtained to reveal the influence of each significant factor, interaction between factors,

and curvature effect. Minitab 17 (Minitab Inc., USA) was used for the factorial design

analysis.

2.7. Experimental Design Using CCD

CCD was employed to elucidate the curvature effect from the factorial design and establish

the quadratic polynomial model. The central point (0) and axial point (α) are added to the

FFD. Experimental conditions in the CCD analysis are summarized in Table 5. Eq. (1) of the

5

quadratic polynomial model was obtained using Minitab 17 (Minitab Inc., USA) [21]:

Y𝑢𝑢 = β0 + β1𝑥𝑥1𝑢𝑢 + β2𝑥𝑥2𝑢𝑢 + β11𝑥𝑥1𝑢𝑢2 + β22𝑥𝑥2𝑢𝑢

2 + β12𝑥𝑥1𝑢𝑢𝑥𝑥2𝑢𝑢 + eu (1)

where Y𝑢𝑢 is the predicted response, β0 is the constant coefficient, 𝑥𝑥1𝑢𝑢 and 𝑥𝑥2𝑢𝑢 are the linear

effect coefficients, 𝑥𝑥1𝑢𝑢2 and 𝑥𝑥2𝑢𝑢

2 are the quadratic effect coefficients, 𝑥𝑥1𝑢𝑢𝑥𝑥2𝑢𝑢 is the interaction

effect coefficient, and eu is an unobserved random error.

2.8. Organic Fouling and Cleaning Experiments

2.8.1. RO dead-end filtration and static cleaning tests

To determine the cleaning efficiencies of the chemical cleaning agents (i.e., NaOH, SDS,

EDTA, and citric acid) on s densely formed organic fouling layer using static cleaning, the

organic fouling and membrane cleaning procedures in a dead-end filtration system are as

follows: 1) filtrate with DI water (membrane conditioning) for 15 min; 2) measure the initial

water flux using DI water for 15 min; 3) filtrate with 1 L of 1 g/L sodium alginate containing

1 mM CaCl2, then, it can gain the 80 mL of permeate water; 4) perform static cleaning with

each chemical agent at 30°C for 1 h; 5) measure the water flux after static cleaning for 15 min,

and 6) calculate the cleaning efficiency using Eq. (2).

𝐹𝐹𝐹𝐹𝑢𝑢𝑥𝑥 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 (𝐹𝐹𝐹𝐹𝐹𝐹, %) = �𝐽𝐽𝑤𝑤𝑤𝑤𝐽𝐽𝑤𝑤𝑤𝑤� × 100 (2)

where Jwi is the initial water flux, and Jwc is the water flux after membrane cleaning.

2.8.2. RO cross-flow filtration and dynamic cleaning test

To find the optimum chemical cleaning efficiency of EDTA on the dense organic fouling layer

in cross-flow mode, the steps used to form the dense fouling layer and the membrane cleaning

6

procedure are as follows: 1) filtrate with DI water for 120 min for membrane compaction, 2)

measure the initial water flux using DI water for 30 min, 3) filtrate with 1 L of 1 g/L sodium

alginate containing 1 mM CaCl2, then, it can gain the 75 mL of permeate water, 4) perform

cross-flow cleaning with EDTA solution using DOE experimental conditions, 5) measure the

water flux after cleaning for 15 min, and 6) calculate the cleaning efficiency using Eq. (2).

3. Results and Discussion

3.1. Static Cleaning Test; Normalized Water Flux (Jw/J0) and Flux Recovery Ratio (FRR)

In this study, four different classes of cleaning agents were used to clean the organic-fouled

RO membrane, including: acid solution (citric acid), alkaline solution (NaOH), metal

chelating agent (EDTA), and surfactant (SDS). To clean the organic-fouled membranes, acid

solutions were used to dissolve metallic oxides and hydroxides and carbonates, whereas

alkaline solutions are applied to remove organic matter by hydrolysis and solubilization.

Metal chelating agents remove divalent cations from the complexed organic molecules.

Surfactants that have both hydrophilic and hydrophobic groups can solubilize

macromolecules by forming micelles around them and help to improve the cleaning efficiency

of the fouled membrane.

Prior to optimizing the chemical cleaning conditions by using DOE, static chemical

cleaning tests were conducted in order to select which chemical agent is most suitable to

remove organic fouling formed through the operation of an RO dead-end filtration system,

reducing the overall number of experimental sets and runs. Fig. 3 shows the changes in water

flux when the RO filtration was operated to form organic fouling on the membrane surface,

static chemical cleaning using four different agents for 1 h, and recovery of the water flux.

From the FRR (%) of each chemical agent (NaOH, citric acid, SDS, and EDTA), 1 wt% of

7

EDTA displayed the best cleaning efficiency of 69.04% (±2.4) compared to the other cleaning

agents. From a previous study [17-19], the efficiency of chemical cleaning depends on

whether the chemical agent can break down the bonding between alginate and Ca2+ when

Ca2+ exists in the organic fouling layer. By the same token, EDTA, a metal chelating agent,

can be combined with divalent cations such as Ca2+, to break down the bonds between

alginate and Ca2+ [12]. Therefore, EDTA was chosen as chemical agent to clean the fouled-

RO membrane, and DOE was then used to optimize the cleaning parameters in this study.

3.2. FFD in RO Cross-flow Filtration System

A multi-factor experiment (FFD) using DOE was carried out at three levels per factor, for

every possible combination of the other factors and their levels using four chemical/physical

cleaning factors on cleaning agent type: chemical concentration, cleaning time, flowrate, and

cleaning temperature (Table 4). By defining the factors and levels, full factorial experiments

that study all paired interactions can be economic and practical to reduce the number of

experimental runs.

Fig. 4 illustrates the significant impact of each factor, their interaction, and center point

with regards to water flux recovery and cleaning efficiency. All independent factors show a

negative slope, which means that the flux recovery at a low-level is higher than at a high-level.

In other words, low-level conditions for all factors have an advantage in restoring the water

flux. In particular, the chemical agent concentration is seen to be one of the most important

variables, showing the largest slope of the factors. A regression analysis was then carried out

to quantitatively identify the factors affecting the flux recovery; a p-value of more than 0.1

was treated as the error-term, and a stepwise method was used. Overall, eight factors

including combinations of concentration, cleaning time, temperature, and flowrate affecting

8

the flux recovery are summarized in Table 6. From the table, the equation model created by

factorial design is deemed unsuitable for determining the optimum chemical cleaning

conditions. Furthermore, the center points of all factors are much higher than the linear line

between the low and high levels. Since a curvature effect was detected in the FFD, CCD was

subsequently performed to model a response variable having curvature.

3.3. CCD in RO Crossflow Filtration System

3.3.1. Regression model equation and analysis by CCD

CCD was used to efficiently estimate the first- and second-order terms and to establish a

model equation by adding center and axial points to previously-performed factorial

experiments (Table 5). The experimental results and the predicted FRR of cleaning by CCD

are shown in Table 7. The following regression equation in coded units was established using

statistical software (Minitab 17, Minitab Inc., USA) to explain the efficiency of chemical

cleaning.

Y𝐹𝐹𝐹𝐹𝐹𝐹 = 85.48 + 0.57A + 0.0410B − 0.348C + 0.0331D − 2.72𝐴𝐴2 − 0.000036𝐷𝐷2 +

0.1561AC − 0.000188BD (3)

where YFRR, as a dependent variable, is the predicted FRR by chemical cleaning, and

independent variables are the concentration of EDTA (A), cleaning time (B), temperature (C),

and flowrate (D). To evaluate the suitability of the regression equation of FRR, Eq. (3) was

verified using a coefficient of determination (R2), a value representing whether a regression

model fits the experimental data. In addition, an adjusted coefficient of determination (adj-R2)

was used to verify the suitability of the model, since the value of R2 tends to increase when

independent variables are added to the model. The values of R2 and adj-R2 at a 90%

9

confidence level were 83.95% and 76.82%, respectively. Therefore, the values predicted using

the equation present a relatively high correlation with the experimental data.

3.3.2. Residual analysis of CCD regression model equation

A residual analysis, which is the difference between experimental and predicted responses,

was performed to verify the model adequacy. Fig. 5 presents the residual plots for error values

in CCD. Model adequacy must have the following conditions: 1) the residual should conform

to a normal distribution; 2) homoscedasticity, i.e., the residual should be randomly scattered

on a center of 0, with no outliers existing on the residual versus fit plot; and 3) independence

and stability, which dictate that the relation between the residual and observation order should

not be a regular pattern.

Fig. 5(a) illustrates whether the probability plot of the residual follows a normal

distribution. Here, the residual was deemed normal since it was linearly distributed between -

5 and +5. In addition, the residual and predicted response are randomly scattered within ±4

(Fig. 5(b)); thus, homoscedasticity was identified. Fig. 5(c) shows that the residual is

irregularly distributed, regardless of the experimental order, i.e., the residual is unaffected by

data order. From these results, the regression model equation of the CCD was determined

such that all elements (normality, homoscedasticity, independence, and stability) of the

residual were satisfied, and was sufficiently confirmed to explain the CCD.

3.3.3. Contour plot and response surface plot of CCD

After running a factorial experiment to determine the significant factors, a response surface

methodology was applied to determine the optimal conditions for chemical cleaning, using a

2D-contour plot and a 3D-response surface plot. The two plots graphically display the effects

10

of the independent variables and their relationships (dependent variables). In Fig. 6(a), the

2D-contour plots show the impact on flux recovery between independent variables; the range

of optimum chemical cleaning conditions for the flux recovery was more than 86%. From

these results, the EDTA concentration range was 0.5 mg/L to 1.1 mg/L, the cleaning time was

20 min to 35 min, the temperature was 20°C to 22°C, and the flowrate was 300 mL/min to

500 mL/min.

In Fig. 6(b), the 3D-response surface plot highlights the potential relationships and

curvature effects between the two independent variables and dependent variable (response).

As such, the 3D-response surface plot can determine the maximum response, which is the

curvature spot of each independent variable. Through 3D-response surface plot, the regression

model equation of the chemical cleaning effect is seen to have a stationary point, and a point

of maximum response confirms its existence in the region of exploration of the independent

variable. From the 3D-response surface plots, the curvature spot of the EDTA concentration

was roughly between 0.5 mg/L and 0.8 mg/L, the cleaning time was close to 20 min, the

temperature was 20°C, and the flowrate was 400 mL/min.

3.3.4. Optimization condition for chemical cleaning of organic fouling

Contour plots could be used to determine the area for the optimization of independent

variables; however, they could not obtain correct values for the optimal conditions for

chemical cleaning. Consequently, in Fig. 7, the response optimizer tool of Minitab 17

(Minitab Inc., USA) was used to find the values for optimal conditions. As results, in a range

of 85-87% flux recovery, the optimal conditions for chemical cleaning are as follows: an

EDTA concentration of 0.68 wt%, cleaning time of 20 min, temperature of 20°C, and flowrate

of 409 mL/min. Under these conditions, the flux recovery is estimated to be ~86.56%.

11

3.4. Verification of Modeling

The optimum conditions of chemical cleaning for RO membrane were derived from the

model equation using DOE. To verify the predicted conditions and cleaning efficiency from

the model equation, comparative experiments were undertaken based on cleaning conditions

recommended from the membrane manufacturer guidelines and DOE analysis. These details

are summarized in Table 8.

In the Table, there are notable differences between the membrane manufacturer

guidelines and results from the DOE analysis, especially with regards to cleaning time and

temperature. The optimal cleaning time and temperature obtained using the DOE analysis

were only 20 min and 20°C, respectively, as opposed to 60 min and 30°C as suggested by the

manufacturer guidelines. The reduction of cleaning time leads to an increase in the

permeation time, and eventually induces a flux increase in the membrane. The low

temperature could also reduce the energy consumption required for chemical cleaning.

Fig. 8 presents the predicted and experimental results for the membrane guidelines

and DOE analysis, which are 83.25% and 85.56%, respectively. Therefore, the cleaning

efficiency predicted using conditions from the DOE analysis was 2.31% higher for the

membrane manufacturer guidelines. In lab-scale experiments, the cleaning efficiencies were

82.82% and 85.07%, respectively, i.e., the cleaning efficiency predicted by DOE was 2.25%

higher than for the membrane manufacturer’s guidelines. Overall, there were minimal

differences between the results predicted by DOE and the lab-scale experiments. In addition,

it is posited here that the conditions predicted by DOE analysis could be closer to the optimal

conditions for chemical cleaning of RO membranes.

Furthermore, the verification of the model equation by lab-scale test indicates that the

12

model equation obtained by DOE is adequate for computationally predicting the cleaning

efficiency of RO membranes with regards to organic fouling.

4. Conclusions and Summary

To optimize the cleaning conditions required to control organic fouling on RO membranes,

factorial designed experiments using DOE were employed using chemical/physical factors

such as cleaning agent type, chemical concentration, cleaning time, flowrate, and cleaning

temperature. The resultant regression equation model governing the effect of operational

parameters on water flux recovery and cleaning efficiency is as follows: Y𝐹𝐹𝐹𝐹𝐹𝐹 = 85.48 +

0.57A + 0.0410B0.348C + 0.0331D − 2.72𝐴𝐴2 − 0.000036𝐷𝐷2 + 0.1561AC − 0.000188BD

(R2 = 83.9 at a 90% confidence level) Based on this regression model, the optimal conditions

for the chemical cleaning of RO membrane are: EDTA concentration of 0.68 wt%, cleaning

time of 20 min, temperature of 20°C, and flowrate of 409 mL/min; the flux recovery is then

estimated to be ~86.6%.

The predicted and experimental results for chemical cleaning conditions obtained

from membrane manufacturer guidelines and the DOE analysis showed that conditions from

DOE analysis were closer to being optimum than those provided by the membrane

manufacturer. Subsequent verification of the model equation via a lab-scale test confirmed

that the model equation obtained by DOE is suitable for predicting the cleaning efficiency of

RO membranes with regards to organic fouling. Overall, the framework suggested in this

study could also provide a promising way for selecting an optimal combination of cleaning

agents and chemical/physical conditions cleaning in other membrane-based filtration systems,

which have inevitably suffered from the formation of organic or biofouling on their

membrane surface.

13

Acknowledgments

This research was supported by a grant (17IFIP-B099786-04) from the Plant Research

Program funded by the Ministry of Land, Infrastructure and Transport of the Korean

government.

References

1. Elimelech M, Phillip WA. The future of seawater desalination: Energy, technology, and the

environment. Science 2011;333:712-717.

2. Kaiya Y, Itoh Y, Fujita K, Takizawa S. Study on fouling materials in the membrane

treatment process for potable water. Desalination 1996;106:71-77.

3. Greenlee LF, Lawler DF, Freeman BD, Marrot B, Moulin P. Reverse osmosis desalination:

Water resources, technology, and today's challenges. Water Resour. 2009;43:2317-2348.

4. Fritzmann C, Löwenberg J, Wintgens T, Melin T. State-of-the-art of reverse osmosis

desalination. Desalination 2007;216:1-76.

5. Hong S, Elimelech M. Chemical and physical aspects of natural organic matter (NOM)

fouling of nanofiltration membranes. J. Membr. Sci. 1997;132:159-181.

6. Lee S, Boo C, Elimelech M, Hong S. Comparison of fouling behavior in forward osmosis

(FO) and reverse osmosis (RO). J. Membr. Sci. 2010;365:34-39.

7. Lee S, Elimelech M. Relating organic fouling of reverse osmosis membranes to

intermolecular adhesion forces. Environ. Sci. Technol. 2006;40:980-987.

8. Liu C, Caothien S, Hayes J, Caothuy T, Otoyo T, Ogawa T. Membrane chemical cleaning:

From art to science. New York: Pall Corporation; 2001. p. 11050.

9. Trägårdh G. Membrane cleaning. Desalination 1989;71:325-335.

14

10. Kim SJ, Oh BS, Yu HW, et al. Foulant characterization and distribution in spiral wound

reverse osmosis membranes from different pressure vessels. Desalination 2015;370:44-52.

11. Garcia-Fayos B, Arnal J, Gimenez A, Alvarez-Blanco S, Sancho M. Static cleaning tests

as the first step to optimize RO membranes cleaning procedure. Desalin. Water Treat.

2015;55:3380-3390.

12. Garcia-Fayos B, Arnal J, Gimenez A, Alvarez-Blanco S, Sancho M. Optimization of

chemical cleaning of a reverse osmosis membrane from a desalination plant by means of

two-step static tests. Desalin. Water Treat. 2015;55:3367-3379.

13. Alayande AB, Kim LH, Kim IS. Cleaning efficacy of hydroxypropyl-beta-cyclodextrin for

biofouling reduction on the reverse osmosis membrane. Biofouling 2016;32:359-370.

14. Shams MB, Mumtaz F, Murali A, Akbar W, Al-Bastaki N. Application of statistical design

to an industrial-scale dead-end ultrafiltration process. Chem. Eng. Technol. 2015;38:1315-

1326.

15. Yi XS, Shi WX, Yu SL, Li XH, Sun N, He C. Factorial design applied to flux decline of

anionic polyacrylamide removal from water by modified polyvinylidene fluoride

ultrafiltration membranes. Desalination 2011;274:7-12.

16. Chen JP, Kim SL, Ting YP. Optimization of membrane physical and chemical cleaning by

a statistically designed approach. J. Membr. Sci. 2003;219:27-45.

17. Mi B, Elimelech M. Organic fouling of forward osmosis membranes: fouling reversibility

and cleaning without chemical reagents. J. Membr. Sci. 2010;348:337-345.

18. Nguyen TV, Pendergast MM, Phong MT, et al. Relating fouling behavior and cake layer

formation of alginic acid to the physiochemical properties of thin film composite and

nanocomposite seawater RO membranes. Desalination 2014;338:1-9.

19. Ramon GZ, Nguyen TV, Hoek EM. Osmosis-assisted cleaning of organic-fouled seawater

15

RO membranes. Chem. Eng. J. 2013;218:173-182.

20. Ang WS, Lee S, Elimelech M. Chemical and physical aspects of cleaning of organic-

fouled reverse osmosis membranes. J. Membr. Sci. 2006;272:198-210.

21. Cochran WG, Cox GM. Experimental design. 2nd ed. New York: Wiley; 1957.

Table 1. Operating Conditions in RO Dead-end Filtration System Operational factors Description Note

Effective membrane area (cm2) 19.6 - Feed solution 1 g/L Sodium alginate (1 L) Contains 1 mM of CaCl2

Dead-end gas pressure (Kgf/cm2) 15.3 ± 0.1 Pressure gas: N2 (99.999%)

Temperature (°C) 20 ± 0.2

Static cleaning time (min) 60 At 20°C

Table 2. Operating Conditions in RO System Operational factors Description Note

Effective membrane area (cm2) 19.3 -

Feed solution DI water Measurement of initial flux and

flux after cleaning Cleaning solution EDTA 0.5 wt% or 2.5 wt%

Hydraulic pressure (bar) 15 ± 0.2 Temperature (°C) 20 ± 0.2

Cleaning time (min) 20 or 80 At 20°C or 40°C Flowrate (mL/min) 100 or 700 -

Table 3. Coded Levels and Independent Factors

Factors Factor details Level details

Low (-1) Center (0) High (+1)

A Concentration of EDTA

(wt%) 0.5 1.5 2.5

B Cleaning Time (min) 20 50 80

C Temperature (°C) 20 30 40

D Flowrate (mL/min) 100 400 700

16

Table 4. Experimental Designs of Flux Recovery Rate Run

order Coded factors Uncoded factors

A B C D A (wt%) B (min) C (°C) D (mL/min) 1 -1 -1 +1 -1 0.5 20 40 100

2 +1 +1 -1 -1 2.5 80 20 100

3 -1 +1 -1 +1 0.5 80 20 700

4 -1 +1 -1 -1 0.5 80 20 100

5 +1 +1 +1 +1 2.5 80 40 700

6 +1 +1 -1 +1 2.5 80 20 700

7 +1 -1 -1 +1 2.5 20 20 700

8 -1 -1 -1 +1 0.5 20 20 700

9 -1 -1 -1 -1 0.5 20 20 100

10 -1 -1 +1 +1 0.5 20 40 700

11 +1 +1 +1 -1 2.5 80 40 100

12 -1 +1 +1 -1 0.5 80 40 100

13 -1 +1 +1 +1 0.5 80 40 700

14 +1 -1 -1 -1 2.5 20 20 100

15 +1 -1 +1 +1 2.5 20 40 700

16 +1 -1 +1 -1 2.5 20 40 100

17 0 0 0 0 1.5 50 30 400

18 0 0 0 0 1.5 50 30 400 * A: concentration of EDTA, B: cleaning time, C: temperature, and D: Flowrate. * Values of Center level are provided by the membrane manufacturer, and values of low and high levels are fixed

by authors to analyze the FFD.

17

Table 5. Experimental Design of Central Composite Design Run

order Coded factors Uncoded factors

A B C D A (wt%) B (min) C (°C) D (mL/min) 1 -1 +1 +1 +1 0.5 80 40 700

2 -1 -1 -1 +1 0.5 20 20 700

3 -1 0 0 0 0.5 50 30 400

4 0 0 0 0 1.5 50 30 400

5 +1 +1 +1 +1 2.5 80 40 700

6 +1 +1 -1 +1 2.5 80 20 700

7 0 0 -1 0 1.5 50 20 400

8 +1 0 0 0 2.5 50 30 400

9 +1 +1 +1 -1 2.5 80 40 100

10 -1 +1 -1 +1 0.5 80 20 700

11 0 -1 0 0 1.5 20 30 400

12 0 0 0 -1 1.5 50 30 100

13 +1 -1 -1 +1 2.5 20 20 700

14 -1 +1 -1 -1 0.5 80 20 100

15 +1 -1 -1 -1 2.5 20 20 100

16 0 0 0 0 1.5 50 30 400

17 +1 -1 +1 -1 2.5 20 40 100

18 -1 -1 -1 -1 0.5 20 20 100

19 0 0 +1 0 1.5 50 40 400

20 0 +1 0 0 1.5 80 30 400

21 0 0 0 0 1.5 50 30 400

22 -1 -1 +1 -1 0.5 20 40 100

23 -1 +1 +1 -1 0.5 80 40 100

24 0 0 0 +1 1.5 50 30 700

25 -1 -1 +1 +1 0.5 20 40 700

26 +1 -1 +1 +1 2.5 20 40 700

27 +1 +1 -1 -1 2.5 80 20 100 * A: concentration of EDTA, B: cleaning time, C: temperature, and D: Flowrate.

18

Table 6. Regression Analysis of Variables after Stepwise Implementation Term Coefficient P-value

Constant 76.709 0.000 Concentration of EDTA (A) -3.306 0.000

Cleaning time (B) -1.236 0.015 Temperature (C) -1.200 0.017

Flowrate (D) -1.617 0.005 Concentration of EDTA (A) vs. Temperature (C) 1.561 0.006

Concentration (A) vs. Flowrate (D) 0.293 0.431 Cleaning time (B) vs. Temperature (C) -0.728 0.087

Cleaning time (B) vs. Flowrate (D) -1.694 0.004 Temperature (C) vs. Flowrate (D) 0.205 0.575

Concentration (A) vs. Temperature (C) vs. Flowrate (D) -0.685 0.102 Cleaning time (B) vs. Temperature (C) vs. Flowrate (D) -0.909 0.045

Ct Pt 8.83 0.000

19

Table 7. Experimental and Predicted Results of Chemical Cleaning of CCD

Standard order

Concentration of EDTA

(wt%)

Cleaning time (min)

Temp. (°C) Flowrate (mL/min)

Flux recovery (%)

Experimental Predicted Difference

1 0.5 20 20 100 83.8415 83.0800 -0.7615 2 2.5 20 20 100 75.0670 74.1440 -0.923 3 0.5 80 20 100 87.3112 84.4120 -2.8992 4 2.5 80 20 100 72.7034 75.4760 2.7726 5 0.5 20 40 100 77.8976 77.6810 -0.2166 6 2.5 20 40 100 74.6631 74.9890 0.3259 7 0.5 80 40 100 78.6486 79.0130 0.3644 8 2.5 80 40 100 76.4706 76.3210 -0.1496 9 0.5 20 20 700 81.0888 83.4040 2.3152 10 2.5 20 20 700 73.6709 74.4680 0.7971 11 0.5 80 20 700 78.8618 77.9680 -0.8938 12 2.5 80 20 700 70.7246 69.0320 -1.6926 13 0.5 20 40 700 80.7163 78.0050 -2.7113 14 2.5 20 40 700 76.6129 75.3130 -1.2999 15 0.5 80 40 700 71.7514 72.5690 0.8176 16 2.5 80 40 700 67.3077 69.8770 2.5693 17 0.5 50 30 400 78.6301 82.7565 4.1264 18 2.5 50 30 400 79.0419 76.9425 -2.0994 19 1.5 20 30 400 80.8219 83.5955 2.7736 20 1.5 80 30 400 82.0896 81.5435 -0.5461 21 1.5 50 20 400 82.8169 83.7080 0.8911 22 1.5 50 40 400 81.5126 81.4310 -0.0816 23 1.5 50 30 100 79.2553 80.8595 1.6042 24 1.5 50 30 700 77.3743 77.7995 0.4252

Center point

1.5 50 30 400 81.5476 82.5695 1.0219

Center point

1.5 50 30 400 86.0058 82.5695 -3.4363

Center point

1.5 50 30 400 85.0704 82.5695 -2.5009

Table 8. Comparison of Optimized Condition by DOE and Membrane Manufacturer Cleaning Guideline

Chemical cleaning condition Membrane manufacturer

guideline Optimized condition by DOE

Concentration of EDTA (wt%) 1.0 0.7

Cleaning time (min) 60 20

Temperature (°C) 30 20

Flowrate (mL/min) 350 400

20

Fig. 1. Schematic diagram of the RO dead-end filtration system.

21

Fig. 2. Schematic diagram of the RO system.

22

Fig. 3. Comparison of normalized water fluxes for static chemical cleaning.

23

Fig. 4. (a) Main effects plots and (b) interaction plots of the flux recovery response (%).

24

Fig. 5. Residual plots of the model for error values in central composite design: (a) normal probability, (b) scatter plot of residual and predicted values, and (c) scatter plot of residual and order.

25

Fig. 6. (a) 2D-contour plots and (b) 3D-response surface plots for the impact on flux recovery between independent variables.

26

Fig. 7. Optimization of chemical cleaning conditions based on response optimizer.

27

Fig. 8. Comparison of flux recovery rate in conditions predicted by DOE analysis and membrane manufacturer cleaning guidelines.

28